摘要
Web serverclusters has become a popular solutionto achieve high performance duetotheirlow price, highreliability,and good scalability. One ofthe key technologiesis “request dispatching”, which is to centrallyacceptallthe incoming HTTPrequests and dispatchthem to the serversinthe clusterto achieve parallelism .The research showsthatthe efficiency ofrequest dispatching is determined by whetherthe serverloads are balanced duringthe dispatching.This paper studies the characteristics ofthe load generated by HTTPrequests,gives anefficient method ofloadsampling,and presentsa novelrequestdispatching algorithm which can predicttheload andbalanceit.The method can be usedto build Web serverclusters ofheterogeneous machines.
Web serverclusters has become a popular solutionto achieve high performance duetotheirlow price, highreliability,and good scalability. One ofthe key technologiesis “request dispatching”, which is to centrallyacceptallthe incoming HTTPrequests and dispatchthem to the serversinthe clusterto achieve parallelism .The research showsthatthe efficiency ofrequest dispatching is determined by whetherthe serverloads are balanced duringthe dispatching.This paper studies the characteristics ofthe load generated by HTTPrequests,gives anefficient method ofloadsampling,and presentsa novelrequestdispatching algorithm which can predicttheload andbalanceit.The method can be usedto build Web serverclusters ofheterogeneous machines.
